Girls' Education at Scale - Review of Evidence - David K. Evans Amina Mendez Acosta Fei Yuan - Co-Impact

Page created by Elmer Rodriguez
 
CONTINUE READING
Girls' Education at Scale - Review of Evidence - David K. Evans Amina Mendez Acosta Fei Yuan - Co-Impact
Girls’ Education
at Scale
Review of Evidence

David K. Evans
Amina Mendez Acosta
Fei Yuan

2021
Girls' Education at Scale - Review of Evidence - David K. Evans Amina Mendez Acosta Fei Yuan - Co-Impact
Research, evaluation and learning are core components of Co-Impact’s work. As we
        embark on a concerted effort to contribute to achieving gender equality at scale in the
        global south, we need to learn about and build on the existing evidence and knowledge.
        As part of our broader learning effort, we commissioned a series of rapid reviews of
        literature by area experts to help us understand major trends as well as new directions
        about what we know works - and doesn’t work - to achieve gender-equitable outcomes
        at scale in the global south. While these reviews serve as a core component of our
        evolving thinking, they do not represent official opinions of Co-Impact.
        Given that these reviews are focused on critical evidence of initiatives that have been
        evaluated at scale, we understand there are experiences and knowledge that may not
        be captured in these documents. We hope to invest in additional reviews in the future
        to cover other areas of inquiry, and also to build on a wider spectrum of evidence and
        perspectives.
        This important work underpins the development of our own research and learning
        strategy, in which we will prioritize the questions and needs of practitioners working to
        achieve gender equitable outcomes, and also to amplify the voices and experiences of
        women, girls, and other marginalized groups. We hope that this evidence and knowledge,
        in turn, will contribute to building the global evidence base.

Co-Impact evidence review series; Review #1                                                         2
Abstract:
            Many educational interventions boost outcomes for girls in settings where girls face
            education discrimination, but which of those interventions are proven to function
            effectively at large scale? In contrast to earlier reviews, this review focuses on large-scale
            programs and policies—those that reach at least 10,000 students—and on final school
            outcomes such as completion and student learning rather than intermediate school
            outcomes such as enrollment and attendance. Programs and policies that have boosted
            access and/or learning at scale across multiple countries include school fee elimination,
            school meals, making schools more accessible, and improving the quality of pedagogy.
            Other interventions, such as providing better sanitation facilities or safe spaces for girls,
            show promising results but either have limited evidence across settings or focus on
            post-educational outcomes (such as income earning) in their evaluations. We discuss
            three aspects of considering evidence-based solutions to local problems—constraints to
            girls’ education, potential solutions, and program costs---as well as lessons for scaling
            programs effectively. If education systems seek to expand and improve girls’ education
            at scale, there are tested tools that have performed effectively in multiple settings, even
            as education leaders, partners, and researchers collaborate to continue innovating and
            testing new programs at scale.
            JEL codes: I21; I24; J16; O15
            Keywords: education; gender; girls’ education; inequality
            Suggested citation: Evans, David K., Amina Mendez Acosta, and Fei Yuan. “Girls’ Education
            at Scale: Review of Evidence.” Co-Impact Research Working Paper. 2021.

1
 Author affiliations: Evans and Mendez Acosta (Center for Global Development); Yuan (Harvard University). This paper was made possible by financial support from
Co-Impact. The authors thank Varja Lipovsek and others on the Co-Impact team for guidance and Radhika Bula for sharing studies from the J-PAL post-primary
education initiative. Pamela Jakiela and Owen Ozier provided useful comments. Author names are listed alphabetically. Corresponding author: Evans, devans@
cgdev.org.

Co-Impact evidence review series; Review #1                                                                                                                   3
Introduction
             Gender equality is a stated objective in much of the world: indeed, the fifth Sustainable
             Development Goal is “achieve gender equality and empower all women and girls”
             (United Nations 2015). Education is a crucial human capital investment that opens the
             door to subsequent economic opportunity. As a result, gender equality in education
             is one crucial step—albeit not the only one—towards achieving gender equality in life
             outcomes more broadly (see Box 1).
             Girls’ education is often touted as one of the best investments in international development
             (Kim 2016). But across low- and middle-income countries, adult women still have less
             education than men. Among young women and men in their early 20s, girls in South
             Asia and Sub-Saharan Africa still have less educational opportunity, whereas in other
             regions, girls have gained more ground (Evans et al. 2021a). These average shifts mask
             important differences across countries, within regions of countries, and across levels
             of schooling. In some parts of the world and at some levels, improved gender equality
             in education means more and better education for girls; in others, it means more and
             better education for boys.
             How to achieve gender equality in education at scale? Evidence on how to expand and
             improve education—including girls’ education—in low- and middle-income countries
             has expanded dramatically in recent years (Cameron et al. 2016; Sabet and Brown 2018;
             World Bank 2018). This review examines evidence from large-scale interventions, usually
             implemented by or in partnership with the government, to improve gender equality in
             education. It also discusses the quality of the evidence, where and how different solutions
             may apply differently, and signals where more evidence may be needed.

             Box 1: Gender equity versus gender equality in education
             Gender equality in education and gender equity in education are different, related
             concepts. Equality may be associated with achieving similar education outcomes for
             boys and for girls. Equity, rather, is associated with “fairness or justice in the provision of
             education” (Espinoza 2007). While the current study focuses on achieving gender equal
             outcomes in education (e.g., gender parity in school completion or learning), that may
             require gender inequality in resources spent. One might also argue that, if women face
             greater challenges than men later in life, gender inequality in education may be needed
             to achieve gender equality in later life outcomes.

             Our results show that programs and policies that have increased school completion or
             boosted learning for girls at scale in areas where girls face educational disadvantage
             include, among others, the elimination of fees or providing scholarships or stipends,
             reducing the distance to school or facilitating travel to school, providing school meals,
             improving the pedagogy of teachers through a range of inputs, and interventions that
             help students receive instruction at their level of learning.2

2
 In this study, we do not distinguish strictly between programs and policies, as policies (such as the elimination of school fees) are virtually always accompanied by
a program (such as providing grants to schools to compensate for lost fees).

Co-Impact evidence review series; Review #1                                                                                                                         4
We also discuss interventions that explicitly address issues faced principally by girls. These
        include sanitation and menstrual health, gender sensitization training, and safe spaces for
        girls. However, most of these interventions either have not been implemented at large
        scale or have not been evaluated with a focus on educational outcomes like improved
        learning and school completion. However, we provide evidence on the interventions
        they do shift (e.g., girls’ mental health and in some cases, their post-school transitions).
        Future, large-scale evaluations of such interventions will allow a better understanding of
        how well such programs can be implemented at scale and whether they shift educational
        outcomes.
        In the discussion section of the paper, we propose a three-step framework—constraints,
        solutions, and costs—for how policymakers and their partners can apply this evidence
        across different contexts. We also discuss lessons for scaling interventions effectively.
        These findings complement those of other, recent reviews related to girls’ education (e.g.,
        Evans and Yuan forthcoming or the girls’ education section of Evans and Mendez Acosta
        2021), but the focus on at-scale programs gives greater salience to programs that have
        been implemented by multiple countries nationwide, such as school fee elimination,
        school construction, or school meals.
        The COVID-19 pandemic introduces large challenges for education overall: for example,
        as of February 2021, schools in South Sudan had been closed for 16 percent of a child’s
        average lifetime schooling careers (Evans et al. 2021b). There are various channels
        by which the pandemic may be particularly harmful for girls: with a higher burden of
        housework while schools are closed, greater risks from possible adolescent pregnancies,
        and discriminatory treatment when resources for education are reduced (Mendez
        Acosta and Evans 2020). Beyond projections, there is little to no data yet available on
        gender differences in learning loss or dropout rates. But insofar as there are differential
        challenges, targeted programs to bring girls back to school and remediate their learning
        may be valuable.

The challenges
        The broad policy challenge
        Inequality in educational achievement is a massive global challenge, but the nature of
        the challenge varies dramatically across settings (Figure 1). For example, in low-income
        countries (like Afghanistan or Mali), boys are more likely to complete primary, lower
        secondary, and upper secondary education than girls. The gap grows with each level
        of education, doubling from a four percentage point gap in primary school (63 percent
        versus 67 percent) to an eight percentage point gap in lower and upper secondary school.

Co-Impact evidence review series; Review #1                                                              5
Figure 1: Gender inequities in education differ at the primary level versus the secondary level,
        and they vary in low-income versus middle-income countries.

        Source: Authors’ construction. The school completion rates for this figure are from the World Development Indicators
        (most recent available year) for primary and lower secondary education, and they are from UNESCO for upper secondary
        education. The data were downloaded in January 2021.

        In lower middle-income countries (like Bolivia or Ghana), girls and boys complete their
        basic education at essentially the same rates. But in upper middle income countries (like
        Malaysia or Mexico), while there is parity in primary school completion (at 97 percent
        – almost every child is completing primary school), girls are five percentage points
        more likely to complete upper secondary school than boys. As countries grow more
        prosperous, gender gaps favoring boys disappear and girls even begin to pull ahead.
        There are exceptions at every level of income. In Madagascar and Burkina Faso, both
        low-income countries, girls are more likely to complete primary and lower secondary
        school than boys. Angola, Benin, and Pakistan are lower-middle income countries with a
        remaining sizeable gender gap favoring boys at the primary level. In some upper-middle
        income countries such as Bulgaria, Gambia, and Guatemala, boys are at least three
        percentage points more likely to finish upper secondary schooling than girls.
        Beyond access, there are also differences in learning outcomes. The World Bank’s
        harmonized test scores show that girls tend to have lower test scores than boys in low-
        income countries, with a concentration in Sub-Saharan Africa. In most middle-income
        countries, including most countries in Latin America, girls outperform boys on exams
        (World Bank 2020b, 19–20). Likewise, beyond differences in the gender gap across

Co-Impact evidence review series; Review #1                                                                                    6
national levels of income, there will be differences in the gender gap across income
        levels within countries, between rural and urban areas (Evans 2019), and across other
        parameters. This array of parameters across which gender inequality can linger explains
        why a middle-income country, despite having achieved gender parity on average, will
        still need to worry about gender equality in education.
        At the same time, the specific challenges that girls (and boys) face change over time. For
        example, in a study of improved sanitation in Indian schools, Adukia (2017) found that
        the gains from sex-specific latrines principally appeared once girls had already reached
        puberty; for younger girls, the presence of a latrine helped, but whether or not it was
        sex-specific mattered less. Concern about sexual violence may be concentrated (albeit
        not limited to) older students. So just as countries will differ in whether they need to
        principally focus on closing gaps favoring boys or girls, the steps will vary based on the
        level of schooling.
        This all reminds us that the challenge of gender equitable education is not a single
        challenge. Countries vary enormously in whether boys or girls are ahead in education
        and, therefore, need special attention and resources. Furthermore, even in countries
        where outcomes are similar, there may be important differences that require distinct
        attention. While this study explores – principally – interventions that have sought to
        improve girls’ education at large scale, no one intervention will apply in all cases.

        The knowledge challenge
        The growth in evidence from evaluations of interventions on how to improve education
        has been dramatic in recent years, with a 15-fold increase in studies between 2000 and
        2016 (Figure 2). Many of these evaluations either focus on girls education or present
        evidence on girls education within the context of a program that benefits both boys
        and girls (Evans and Yuan forthcoming). Beyond methodological differences (e.g.,
        experimental versus quasi-experimental evaluations), this evidence includes evaluations
        of various types of programs. Policymakers and partners can learn different things from
        each type of program evaluation.

Co-Impact evidence review series; Review #1                                                          7
Figure 2: The growth in cumulative experimental and quasi-experimental education impact
        evaluations over time

        Source: Authors’ construction. Adapted from Figure S4.1 from World Bank (2018).

        One way to categorize the programs evaluated is based on implementer and scale,
        yielding four types: (1) pilot interventions implemented by non-government actors, (2)
        pilot interventions implemented by government actors, (3) large-scale interventions
        implemented by non-government actors, and (4) large-scale interventions implemented
        by government actors (Figure 3). For the sake of this paper, interventions that reach at
        least 10,000 students or were implemented nationwide will be considered “large-scale.”
        (In some countries, even nationwide implementation may not reach 10,000 students.)
        Obviously that number is ultimately an arbitrary one, but it does allow a distinction
        between truly small programs and those that require more management infrastructure
        and resources to implement.
        An example of the first type of study—pilot interventions implemented by non-
        government actors—might be the enrollment of about 2,000 children in alternative
        schools established by non-government organization (NGO) in Guinea-Bissau (Fazzio et
        al. 2020). That intervention led to a six-fold increase in girls’ test scores. In that context,
        government provision of education is very limited, and this study demonstrates that a
        holistic intervention—providing non-government schools with custom-designed teacher
        training materials and classroom materials for students and teachers, together with
        monitoring and evaluation of teachers and community outreach—can deliver dramatic
        gains in learning in an area with historically low learning levels. Another example of such
        a program evaluation examines the impact of setting up parent-teacher conferences in a
        study of about 4,000 students in Bangladesh (Islam 2019). The program was implemented
        by a local NGO, and it more than doubled girls’ test scores by the end of two years. Both
        studies provide promising interventions to increase girls’ learning, but neither tells us
        whether it would be possible to implement such a program at scale. Evaluations of this

Co-Impact evidence review series; Review #1                                                               8
style of intervention can be designed to inform scale up, but how well the program will
             actually work with thousands more students remains a challenge to know with certainty
             (Banerjee et al. 2017).

             Figure 3: What policy makers and partners learn from different types of program evaluations

              Scale

                                                            Small                             Large
                Implementation mechanism

                                           government

                                                            Does the theory behind the        Is it possible to reach large numbers of
                                              Non-

                                                            intervention apply in real life   children and youth with the program?
                                                            and in this setting?

                                                            Can existing structures
                                               Government

                                                                                              Is it possible to reach large numbers
                                                            implement this program under      of children and youth through existing
                                                            careful supervision?              structures?

             Source: Authors’ construction

             The second type of study implements an intervention at relatively small scale but using
             government systems. For example, the Government of South Africa implemented a
             randomized controlled trial (RCT) in 180 public primary schools, comparing the provision
             of traditional teacher professional development with the more interactive, on-site
             coaching. The coaching boosted girls’ test scores four times as much as the traditional
             training (Cilliers et al. 2019).3 Because this was implemented through government
             channels, we can be more confident that it is possible to implement using teachers
             who have been recruited and are remunerated and managed through the government
             system. This points to a greater confidence that the intervention may scale (Gove et al.
             2017).
             However, even within government systems, implementing a program on a large scale
             still poses new challenges (Anderson 2018). First, the quality of implementers may suffer.
             Another government-implemented pilot impact evaluation in South Africa provided
             teachers with virtual coaching, i.e., a coach who reached out to teachers via tablet.
             Because the coach did not travel to schools, one coach was able to provide virtual
             support to many schools (Cilliers et al. 2020a). But if such a program were scaled up
             nationally, more coaches would be needed. Would it be possible to find many coaches
             of similar quality of the pilot coach? In some places, maybe not. Second, the quality of
             supervision may suffer. Providing careful supervision to a pilot with a dozen or even a
             hundred schools is a different endeavor than providing careful supervision to thousands
             of schools. Third, in pilot programs it may be relatively easy to adjust the program as
             new challenges arise. Large-scale programs lose some of that quick flexibility. Fourth,
             programs often change models at scale to reduce costs or because of political pressures.

3
 The separated boy-girl impacts are not reported in the study, but they were communicated to us by the authors. For girls, the impact of coaching on literacy was
0.15 standard deviations (p=0.10) and the impact of traditional training was 0.04 SDs.

Co-Impact evidence review series; Review #1                                                                                                                    9
A literacy program in Uganda was highly effective at boosting student learning when
             implemented by an NGO but was completely ineffective when a reduced cost version
             was implemented by civil servants (Kerwin and Thornton forthcoming). A program
             putting teachers on short-term contracts led to learning gains when implemented by
             an NGO, but its design was compromised when implemented at scale due to political
             pressures (Duflo et al. 2015; Bold et al. 2018).
             Because of these and other challenges, large-scale programs often report smaller impacts
             than pilot programs, as has been demonstrated both in teacher coaching programs in
             high-income countries (Kraft and Blazar 2018) and across a variety of education programs
             in low- and middle-income countries (Evans and Yuan 2020).4 Thus, rigorous evaluation
             evidence of programs at scale is uniquely valuable.
             Studies of the third or fourth type – implemented at scale, either in government systems
             or through NGOs that have the capacity to work at large scale – provide the most
             direct evidence of effective, at-scale interventions. Most of these evaluations are quasi-
             experimental. For example, an evaluation of a government program providing funds for
             bicycles to 160,000 secondary school girls in India (Bihar state) compares the girls who
             received bicycles to girls in a neighboring state and to local boys, who were not eligible
             for the program, a method called “triple differences” (Muralidharan and Prakash 2017).
             The aforementioned study of latrine construction in Indian schools compares changes in
             outcomes among students attending schools where latrines were built through a large-
             scale government program (Adukia 2017). These evaluations show that it is possible to
             implement the program at scale and still achieve significant impact.
             While it is possible to learn from all these classes of evaluations, the focus of this
             study will be on studies that fall into the latter two categories, especially those that are
             implemented by government at scale.5 Most education is provided by the public sector:
             across low- and middle-income countries, less than 20 percent of primary education and
             less than 30 percent of secondary education is provided privately (World Bank 2020c).6
             As a result, interventions at large scale may be most sustainable if implemented through
             public sector mechanisms. NGOs that achieve results at large scale are also a key source
             of information for both policy makers and donors.

             The methods used for this review
             This is a narrative, rapid review of evidence on large-scale efforts to improve girls’
             education at scale. The research team examined a long list of evaluations and identified
             those that reports impacts for girls and were implemented at large scale (at least 10,000
             beneficiaries or nationwide implementation). The evaluations were drawn from various
             sources, including previous reviews of evidence on girls’ education that included
             both small and large-scale studies (Evans and Yuan forthcoming; Sperling et al. 2016;
             Awasom et al. 2020), the Millions Learning initiative (Robinson et al. 2019), the J-PAL

4
  DellaVigna and Linos (2021) highlight that this pattern may be driven by “publication bias,” in which statistically significant results are more likely to be
submitted and accepted for publication in academic journals (see section 4.3.1 of their study). Studies with smaller samples have less statistical power to
estimate an effect, so smaller studies will on average require larger effects to show statistical significance.
5
  Previous reviews have examined the full array of evidence. See Evans and Yuan (forthcoming) for one example.
6
  The numbers are 19 percent for primary and 28 percent for secondary. This is the low- and middle-income country aggregate provided for 2019, the most
recent year for which data are available. This represents an increase from ten years previously, when the numbers were 15 percent and 23 percent for primary
and secondary.

Co-Impact evidence review series; Review #1                                                                                                                   10
post-primary education initiative, the “Learning @ Scale” initiative (Stern et al. 2020), and
        reviews on individual education topics (e.g., Read and Atinc 2017). The team included
        both evaluations that target girls specifically and evaluations that target both boys and
        girls but that report impacts separately for girls (see Box 2), or studies for which an earlier
        review identified gender-separated effects that were not reported in the original studies
        (Evans and Yuan forthcoming).

        Box 2: Gender targeted versus nontargeted interventions
        Interventions to improve school completion and learning for girls may include either
        interventions that are targeted specifically to girls (e.g., a girls’ scholarship) or general
        interventions. Previous research demonstrates that in some circumstances, general
        interventions deliver comparable gains to targeted interventions (Evans and Yuan
        forthcoming). The choice of targeted or general interventions will depend on the principal
        obstacles girls face and whether those obstacles are principally felt by girls or not.

        The principal outcomes of interest were those nearer the end of the education
        production process: learning and school completion. Studies with other outcomes, such
        as enrollment or attendance, were included principally insofar as they were instrumental
        to learning or to school completion, or in cases where other studies of similar programs
        established impacts on learning or completion. Subsequent life outcomes—e.g., income,
        fertility, or employment—were included in the relatively rare cases that they were available.
        We reviewed this collection of studies and synthesized and summarized the findings.
        While this review is focused around improving gender equality in education, we include
        discussion of studies that impressively boost educational outcomes for girls in areas
        where girls remain disadvantaged in school—particularly Sub-Saharan Africa and South
        Asia (Evans et al. 2021a)—even in cases where both girls’ and boys’ education increase.
        In those cases, similar absolute gains may reduce inequality if girls begin at a lower level:
        i.e., an increase in primary school completion of ten percentage points represents a
        higher percentage increase for girls if they begin at a lower level of completion. Even in
        cases where the percentage gains are similar, sizeable and significant gains to the quality
        and accessibility of girls’ education is likely to benefit girls regardless of impacts on boys.
        Our discussion of the effects is largely qualitative. In this rapid review, we do not
        standardize effect sizes across interventions. Standard deviations in outcomes, while
        useful, can vary widely due to factors separate from the impact of the intervention such
        as the underlying distribution of initial value of the outcome in the study population, as
        well as – in the case of measuring learning – differences in the difficulty of changing the
        specific skills measured or different test designs (Evans and Yuan 2020).

Co-Impact evidence review series; Review #1                                                               11
The solutions
        Make school cheaper
        Fee elimination and scholarships
        Many studies demonstrate that eliminating school fees or providing scholarships can
        dramatically increase school completion as well as learning outcomes. This applies in
        both primary and secondary school. While most countries have officially eliminated
        school fees at the primary level and some have eliminated school fees at the secondary
        levels, in practice, families are often required to pay some sort of contributions to the
        school, above and beyond the cost of school materials and transportation and the
        opportunity cost of sending children and youth to school (Williams et al. 2015).
        At the secondary school level, providing scholarships for youth in Ghana who had
        passed the secondary school entrance exam but who did not have the resources to
        pay—so keep in mind that this is a select group of students—led to more than a 60
        percent increase in senior secondary school completion for girls (an increase of 25
        percentage points relative to 40 percentage points in the comparison group). Girls were
        also much more likely to be enrolled in tertiary, although few were enrolled to begin
        with: that increased from 8 percent to 12 percent. The scholarships also led to a range
        of other positive impacts: better test scores in both reading and mathematics, better
        national political knowledge, media engagement, and a higher likelihood of having a
        bank account. Girls even had fewer pregnancies (Duflo et al. 2019). Eliminating school
        fees for secondary school girls in the Gambia increased the number of girls taking the
        high school exit exam (one proxy for completion) by 55 percent (Blimpo et al. 2019).
        Likewise, providing vouchers to cover the cost of private secondary school in Colombia
        – a program that reached 125,000 children – increased both test scores and secondary
        school completion rates, at comparable rates for both girls and boys (Angrist et al. 2006).
        Another scholarship program – for more than 100,000 girls in upper primary and lower
        secondary grades in the Democratic Republic of the Congo – boosted both reading and
        mathematics scores (Randall and Garcia 2020).
        Another program that paid school fees – this time in Tanzania – also covered other,
        informal costs for tens of thousands of secondary school girls who had been identified
        by their communities as highly vulnerable. The program also provided other benefits,
        such as textbooks and life skills training. Beneficiary girls had much higher test scores
        than girls at comparison schools. Less poor girls who attended beneficiary schools but
        did not receive financial support also had higher test scores, as did boys, suggesting
        a positive spillover effect. Girls who had their fees paid were 25 percent less likely to
        drop out of high school (Sabates et al. 2020). The evaluation of this program matched
        beneficiary girls with girls at other schools based on observable characteristics, so we
        are slightly less confident of the causal impact claim—i.e., it is possible that students in
        beneficiary schools were different than their comparators in ways we do not observe but
        that affect educational outcomes. Still, the results are consistent with evidence elsewhere
        that eliminating fees can boost learning and reduce dropout.

Co-Impact evidence review series; Review #1                                                            12
Feed children at school
        There is a long history of evaluation evidence demonstrating that school meals boost
        enrollment at school. A large-scale school meals program in Pakistan – reaching
        hundreds of thousands of girls – reduced one measure of malnutrition (wasting) by
        almost half and boosted school enrollment by forty percent (Pappas et al. 2008). There
        is some evidence that children across 32 African countries benefiting from a World Food
        Program school feeding initiative were enrolled in school at higher rates, with 27 percent
        higher gains for girls than for boys (Gelli et al. 2007).
        A new generation of evidence demonstrates that school meals boost student learning
        as well. In Ghana, a large-scale school feeding program for which funding is now
        integrated into the government budget was evaluated via randomized controlled trial.
        After two years, both math and literacy scores rose for all children on average, but the
        largest impacts were for girls and for children in poverty. In other words, school feeding
        boosted learning especially for girls (Aurino et al. 2020). Likewise, a large-scale midday
        meal program in India led to improved test scores in both math and reading: in that case,
        girls and boys benefitted equally, as did poorer and less poor children (Chakraborty and
        Jayaraman 2019).

        Cash transfers
        Cash transfers are a widely used tool to achieve multiple ends: often the primary goal
        is that of a social safety net—ensuring that low-income households have money for
        essential needs—but they are often paired with further objectives related to health and
        education, either explicitly though conditions that households must fulfill to receive
        the transfers or more subtly through labeling and encouragement (Benhassine et al.
        2015). Many of these programs have been implemented at scale, and there are many
        variations, including those that combine a transfer with the direct payment of school
        fees, as in Bangladesh (Schurmann 2009). There have been many evaluations of cash
        transfer programs on education, and most of those (nearly two-thirds) do report impacts
        separately for girls (Evans et al. 2020). However, one recent, fairly comprehensive review
        identified only a handful of studies that reported test scores or grade completion (Baird
        et al. 2014). That review reports consistent positive impacts on school enrollment—with
        higher impacts for conditional programs—but small impacts on test scores. Indeed, a
        recent report rates cash transfers as a “bad buy” if the objective is to boost learning
        (Global Education Evidence Advisory Panel 2020). That said, ensuring that youth are in
        school is a likely precondition for learning and completion outcomes, so while cash
        transfers are not an effective instrument for learning alone, they may be crucial to making
        sure the most vulnerable children have the opportunity to reap learning gains from other
        interventions.

        Make school more physically accessible
        Two classes of interventions to make schools more accessible—constructing schools
        and providing transportation—have been implemented at scale with success and have
        shown positive impacts. Indeed, the Global Education Evidence Advisory Panel’s Smart

Co-Impact evidence review series; Review #1                                                           13
Buys report identifies reducing travel times to schools as one of its “good buys” (its
        second best rating, after “great buys”) for boosting learning in general, citing school
        construction and school transportation as two tested instruments (Global Education
        Evidence Advisory Panel 2020). The benefits are experienced disproportionately by girls.
        Distance to school is a major constraint for many girls, especially at the secondary school
        level. For example, teenage girls in India who live 15 kilometers from a school have been
        more than one-third less likely to attend than those who live near a school (Muralidharan
        and Prakash 2017).
        One solution to that challenge is to build schools close to girls. Several interventions
        that have employed this approach have constructed schools with the needs of girls
        in mind (i.e., “girl-friendly” schools). A program that constructed primary schools in
        Burkina Faso benefited many thousands of children (the precise number is unclear): after
        2.5 years, enrollment and test scores rose, and both of these effects were greater for
        girls than for boys (Kazianga et al. 2013). After ten years, impacts on test scores and
        enrollment remained although they were smaller, which may be unsurprising because
        many comparison villages also had some sort of school. However, primary completion
        rates for girls were more than double in beneficiary villages than in comparison villages
        (23 percent versus 9 percent); they were also much higher for boys (39 percent versus 30
        percent). Marriage rates for girls were also much lower in beneficiary villages (33 percent
        versus 39 percent). The schools all had separate latrines for boys and girls and a water
        source. Importantly, the program did not merely construct schools: it also provided
        school meals, books, and an information campaign to parents on the importance of
        education (Davis et al. 2016). A slightly smaller scale program in Niger boosted student
        test scores, but only for girls (Bagby et al. 2016).
        We know something about the long-run impacts of school construction. In Indonesia,
        the government implemented a massive school construction program: more than 61,000
        schools were constructed between 1973 and 1979. Women were more likely to complete
        primary school (by 4 percentage points) as a result, and their children were more likely
        to complete secondary school and even tertiary education, with slightly larger effects on
        their daughters than on their sons (Akresh et al. 2019).
        A related intervention extends existing schools. In several countries of Latin America
        (Brazil, Colombia, Ecuador, Honduras, and Nicaragua), a program provided alternative
        lower- and upper-secondary education to youth who otherwise would not have access.
        The program is “alternative” in the sense that it seeks to integrate academic learning with
        practice livelihood experience, and it has reached some 300,000 students (Kwauk and
        Robinson 2016). An evaluation of the program in Honduras demonstrated higher test
        scores at lower cost than traditional schools (McEwan et al. 2015).
        The second involves making transportation to school more accessible. In India, a program
        provided cash transfers intended for the purchase of bicycles to more than 150,000 girls.
        School principals then provided receipts demonstrating that the cash had been used to
        purchase bicycles. The impacts of the program are striking: the gender gap in secondary
        school fell by 40 percent, and the girls who passed the high-stakes secondary school
        exam rose by 12 percent (Muralidharan and Prakash 2017). A much smaller version of the
        program (benefiting several thousand girls) distributed bicycles to younger schoolgirls

Co-Impact evidence review series; Review #1                                                           14
(in upper primary school); the evaluation found the program reduced absenteeism,
        commute time, and teasing but had less dramatic effects otherwise (Fiala et al. 2020).

        Teach better
        General improvements in the quality of instruction
        In Kenya, a multi-faceted literacy and numeracy program was implemented through
        government systems. It included detailed teachers’ guides, training for teachers and
        head teachers, teacher coaching, and literacy and math books for every student. The
        pilot program—evaluated via a randomized controlled trial—led to impressive literacy
        gains in the early years of primary school, with comparable impacts for girls and boys.
        Impacts for mathematics were more modest but still positive (Piper and Mugenda 2014).
        When the program was scaled up nationwide (an initiative called Tusome), literacy gains
        were still positive and sizeable, even though some aspects of implementation were not
        as well done as they were at the pilot. For example, teachers received some coaching in
        the scaled up version but less than planned (Piper et al. 2018). Gains were slightly bigger
        for girls than for boys on several of the literacy measures (Fraudenberger and Davis 2017).
        A related program, implemented at large scale in Pakistan and evaluated using a quasi-
        experimental program, also provided detailed teachers’ guides and reading materials
        for students, among other activities (Chemonics International Inc. 2019). Beneficiary
        children boosted their reading more than non-beneficiary children, with larger gains for
        girls (Management Systems International 2018).

        Target children who fall behind
        In India, a remedial reading program implemented by Pratham, a large NGO, reaches
        hundreds of thousands of students. Hiring young women to teach students who are
        falling behind in their basic numeracy and literacy skills led to significant gains for both
        girls and boys (Banerjee et al. 2007). Pratham has implemented several variations on
        programs to bring learners who are falling behind up to speed, including intensive
        summer reading camps or using one hour of the school day to group learners by ability
        rather than official grade level (Banerjee et al. 2017).
        In Ghana, the government implemented a related program with more than 80,000
        students without NGO-support (beyond a visit to Pratham in India to see their programs
        in action). Schools hired teaching assistants, many of whom had no previous experience
        with teaching. In one model, the assistants pulled remedial learners out of class for part
        of the school day for special attention. In a second model, the assistants provided the
        same attention but after school hours. In a third, assistants worked with half of a class for
        part of the school day, effectively just reducing class size and allowing more specialized
        attention per student. In a fourth model, there were no teaching assistants, but teachers
        were trained to focus their teaching at whatever learning level students were at, rather
        than being bound by the curriculum for their grade. All four models led to significant
        gains in student learning, and three of the four (all except the third) were equally cost-
        effective. Gains were higher for girls than for boys (Duflo et al. 2020).

Co-Impact evidence review series; Review #1                                                             15
What about other teacher policies?
             Teacher incentives in India, implemented in schools covering more than 20,000
             students, led to sizeable, significant gains in language and especially math scores, with
             no significant differences for girls and boys (Muralidharan and Sundararaman 2011).
             In another teacher performance pay intervention in Tanzania that reached more than
             120,000 students, students in schools that received a combination of school grants and
             teacher performance incentives saw the largest test score gains, and the gains for girls
             were significantly larger than those for boys (Mbiti et al. 2019).7
             A large-scale teacher professional development in China had no impact on student
             learning for girls (or for boys), arguably because the training was too theoretical (Loyalka
             et al. 2019). More broadly, at-scale teacher professional development programs tend not
             to incorporate the elements that are associated with the best student learning outcomes
             in smaller programs (Popova et al. forthcoming). A multi-faceted educational program in
             Tanzania, evaluated with a quasi-experimental design, led to significant gains in literacy
             and numeracy in the early grades of primary school, with the biggest gains for girls. While
             attribution across components is difficult, the authors hypothesize that the teacher
             professional development played a major role, as it was implemented effectively and led
             to increase use of teaching aids, boosted teacher confidence and motivation (Ruddle
             and Rawle 2020).

             Deploy effective forms of education technology to improve instruction
             Many education systems invest extensively in education technology (or ed-tech),
             although reviews have found decidedly mixed impacts of education technology
             investments on student learning outcomes (Bulman and Fairlie 2016; Escueta et al. 2020;
             Evans and Mendez Acosta 2021; Rodriguez-Segura 2020).8 The heterogeneity of impacts
             likely derives from the fact that technology plays many roles in education: technology
             can be used to improve the quality of instruction, to nudge parents or students, or for
             self-led learning. Some interventions to improve the quality of instruction have been
             successful. Many of these programs have been implemented at large scale, although the
             evaluations often study a smaller sample of students. For example, providing children
             with technology-aided after school instruction in urban India lead to large gains in math
             and language scores that were comparably sized for boys and girls. The evaluation
             sample was fewer than one thousand students, but the software has been used by
             more than hundreds of thousands of students (Muralidharan et al. 2019). However, this
             presents a conundrum: a small evaluation may have much more controlled conditions
             than large-scale use of a software. In this case, a larger, in-school version of the program
             was implemented for one period per day and still delivered significant positive learning
             impacts, albeit smaller than those in the pilot, after school program (Muralidharan and
             Singh 2019). Likewise, a large-scale computer-assisted learning platform in Uruguay that
             has reached at least half of all students in third through sixth grade delivered significant

7
  Another class of teacher incentives provides additional financing to teachers who work in rural or otherwise disadvantages schools. While those incentive
programs, implemented at scale, tend to be effective in leading teachers to relocate, they did not boost girls’ test scores in Zambia (Chelwa et al. 2019), although
they did, particularly for higher income girls, in the Gambia (Pugatch and Schroeder 2018).
8
  Some high-performing education systems in high-income countries have actually reduced investments in education technology relative to education systems in
other high-income countries (Ripley 2014).

Co-Impact evidence review series; Review #1                                                                                                                      16
learning gains of comparable size for both girls and boys (Perera and Aboal 2019). Both
             at-scale computer-assisted learning programs delivered similar results: roughly 0.2
             standard deviations of mathematics learning, which amounts to a significant gain relative
             to a year’s learning (Evans and Yuan 2019).
             There is less evidence on educational television, although there are several ongoing
             studies of educational television and radio in the context of the COVID crisis (World
             Bank 2020a). The existing evidence suggests that high-quality educational television
             programming can boost cognitive outcomes for girls in particular, although that evidence
             is from younger children (Mares and Pan 2013). That said, across African countries, even
             with schools closed during the pandemic, relatively few households have access to ed-
             tech products (Crawfurd 2020).
             It is with some reserve that we include ed-tech among the solutions, since simply
             providing education technology does not inherently promise any gains. Indeed, several
             large-scale programs have provided computing equipment and had no impact on
             student learning or other outcomes. This result is important because scaling hardware
             programs, while costly, may appear logistically straightforward. But providing computers
             to more than 6,000 schools in Colombia had no impact on student learning for girls or
             for boys (Linden and Barrera-Osorio 2009). Providing laptops (more than 1.6 million of
             them!) to children in Uruguay had no impact on girls’ learning outcomes in either the
             short run or the longer run (Yanguas 2020); a similar program that delivered tens of
             thousands of laptops in Peru had no impact on girls’ learning nor on students’ dropout
             rates (Cristia et al. 2014; 2017). Providing laptops or desktops may be a necessary step
             in providing technology-assisted learning, but programs that do the first without the
             second are poised for failure.
             Many other technological innovations have been either implemented only at smaller
             scale, with far fewer than 10,000 students (e.g., Berlinski et al. 2016 or Duflo et al. 2012).
             In other cases, ed-tech interventions have been implemented at large scale but lack
             serious evaluation on student outcomes, e.g., several large-scale education technology
             platforms implemented in India (Dhar et al. 2016; Bajpai et al. 2019).

             Gender focused interventions
             While many obstacles affect girls and boys differentially, several classes of interventions
             seek explicitly to address obstacles felt principally by girls. Many of these interventions
             are not focused on academic outcomes, and so do not report school completion or
             learning outcomes,9 although there are exceptions. But a lack of evidence on those
             outcomes does not mean these are not important investments. It may mean that the
             primary motivations for investing in some of these programs are not to boost learning
             outcomes or school completion but rather to protect girls, to boost their overall well-
             being, or to prepare them for life after school.

9
 One example of an evaluation that did not report learning or completion outcomes is the Power to Lead program, which reached tens of thousands of adoles-
cent and pre-adolescent girls across six countries. A mostly qualitative evaluation found positive impacts on measures of leadership skills, leadership action, and
self-confidence (Miske Witt & Associates Inc. 2011).

Co-Impact evidence review series; Review #1                                                                                                                      17
Sanitation and menstrual health products
        Toilets for girls often arise in discussions of gender-equitable education. Indeed, a
        national school latrine construction program in India increased enrollment for girls. For
        the youngest girls, any latrine boosted enrollment. For girls who had reached puberty
        (i.e., upper primary), only sex-specific enrollment boosted outcomes. While the study did
        not measure results on completion, the enrollment results are strong and enduring three
        years after construction. Students did not perform better on direct tests of their learning,
        but girls (and boys as well) both sat for and passed their official school exams at higher
        rates (Adukia 2017).
        Two reasons that sex-specific latrines often come up in discussions is because of the
        fear that girls will either miss school during menstruation or that they might experience
        verbal or physical harassment. On the former point, absenteeism due to menstruation
        may not be a major problem: in Nepal, researchers estimated that girls on average
        missed less than half a day over the course of the school year (Oster and Thornton 2011).
        Menstruation likewise does not increase the gender gap in education in Malawi (Oster
        and Thornton 2011), and in Kenya, girls reaching the age of menarche were not more
        likely to be absent than boys—except for absences due to school transfers (Benshaul-
        Tolonen et al. 2019). In the Kenya study, providing sanitary pads (which are girls’ choice)
        did reduce absenteeism, although those results did not hold up if one excludes girls who
        had transferred away from the study schools. In both Kenya and Nepal, providing an
        alternative, less familiar technology—a menstrual cup—did not affect absenteeism. (The
        cup is much cheaper, but an intervention cannot be cost-effective if it is not effective!)
        Regardless, these are both relatively small studies, with a few hundred students in
        Nepal and a few thousand in Kenya. However, looking beyond absenteeism, providing
        either pads or a cup in Kenya significantly improved emotional and social well-being
        (Benshaul-Tolonen et al. 2020b); and in Tanzania, having insufficient menstrual materials
        was associated with more teasing of girls about their periods (Benshaul-Tolonen et al.
        2020a). On net, the evidence for providing menstrual hygiene materials on a purely
        instrumental basis (getting girls to school completion or higher test scores) is still weak,
        although the emotional well-being of female students is another important, worthy end
        in itself.

        Gender-sensitization training
        When discussing gender equality in education, a class of program that often arises
        in discussion is some sort of training to increase the sensitivity of teachers, school
        managers, or students to gender issues. To date, there is limited evidence evaluating
        such programs at large scale or on outcomes such as school completion or student
        learning. For example, one program evaluated the impact of 2.5 years of classroom
        discussions on the topic of gender equality among several thousand sixth and seventh
        graders in India. (The evaluation sample was about 14,000, with just under half of those
        students receiving the program.) While that is not a small-scale program, it also isn’t at
        a massive scale, and the outcomes measured are reported views on gender norms and
        some self-reported behaviors (e.g., boys reported a higher likelihood of doing chores).
        The program tested neither student learning (beyond on gender attitudes) nor school
        completion (Dhar et al. 2020).

Co-Impact evidence review series; Review #1                                                            18
Another program that functionally acted as a gender sensitization program was the Power
        to Lead program, which provided training leadership skills for girls across six countries. In
        practice, more than 30 percent of participants were boys, and these reported improved
        understanding of gendered social norms (Baric 2013).
        The fact that neither of these studies measured school completion or student academic
        learning is not a critique: those were not the principal objectives of the programs, and
        reducing sexism is a valuable objective in its own right. But it may also have instrumental
        value, increasing completion or student learning—for example, if girls are able to focus
        more on their studies in a more gender sensitive environment (i.e., with less teasing or
        harassment). Further research will be required to see if this is in fact the case.

        Safe spaces for girls
        Creating places where female students can engage without boys or men is sometimes
        proposed as a useful intervention (Megevand and Marchesini 2020). Indeed, evidence
        on this type of intervention is promising, although the outcomes measure is not school
        completion or student test scores. For example, a program that formed clubs for more
        than 50,000 adolescent girls in Uganda and provided vocational training and information
        about sex, reproduction, and marriage, led to reduced adolescent pregnancy and more
        engagement in income-generating activities four years later (Bandiera et al. 2020a).
        However, an adaptation of the same program in Tanzania had no impact on similar
        outcomes (Buehren et al. 2017).
        A program forming girls’ clubs in Sierra Leone—with a smaller sample of 150 beneficiary
        villages—was interrupted by school closures due to the 2014/2015 Ebola epidemic; but
        five years later, girls in communities where girls’ clubs had initially been formed were
        much less likely to have experienced an adolescent pregnancy and much more likely
        to have re-enrolled in school after the epidemic (Bandiera et al. 2020b). These are
        important outcomes beyond education; the impact of these types of programs on purely
        educational outcomes are less well known at a large scale. A small scale, government
        implemented program in 20 low-performing secondary schools in Trinidad and Tobago
        converted co-education schools to single sex schools: girls subsequently performed
        better on secondary school completion exams, and both boys and girls took more
        advance coursework (Jackson 2021).

        Other interventions
        Effective interventions with less evidence behind them
        Some interventions have proven effective at scale but have been rigorously evaluated in
        only one or two settings, so policymakers and partners may feel less confident that their
        success can be replicated elsewhere. For example, a program that provided eyeglasses
        to almost 30,000 students in China boosted test scores equivalent to nearly a full year
        of business-as-usual schooling for students with poor vision, benefiting both girls and
        boys, although one-third of girls refused eyeglasses while only one-fourth of boys did
        (Glewwe et al. 2016). Likewise, a program in India – implemented by the NGO Pratham
        (like some of the literacy programs above) – reached a large sample of mothers to either

Co-Impact evidence review series; Review #1                                                             19
provide literacy for mothers, train mothers in how to boost their children’s literacy, or
        both. All three variations of the program had positive (but modest) impacts on children’s
        mathematical performance (Banerji et al. 2017). Of course, providing eyeglasses to
        children who need them is a worthy and important objective, as is making sure that
        mothers can support their children at home. But the question is not the worthiness of
        the objective but rather how well such programs can be implemented at scale and the
        gains relative to other programs in terms of outcomes of interest.

        Areas with more limited, mixed evidence
        School accountability, often involving publicizing either resources flowing to schools
        or student performance in schools, is an important area. In Uganda, using newspapers
        to publicize the amount of funds that schools would be receiving from the central
        government boosted student learning outcomes, with apparently larger effects for girls
        (Reinikka and Svensson 2011). But programs publicizing results – at scale – have been less
        effective. For example, a program in India that facilitated meetings to discuss education
        and – in some cases – invited community members to create “report cards” on learning
        status in the community had no impact on learning outcomes for girls (or boys) (Banerjee
        et al. 2010). In Tanzania, a nationwide, low-stakes accountability program published
        school rankings: while it boosted learning outcomes in the poorest performing schools,
        it also led those schools to exclude students – both girls and boys, to equal degrees –
        from their last year of schooling, an unfortunate, unintended consequence (Cilliers et al.
        2020b). Many school accountability programs reached large numbers of students do not
        report impacts separately for girls (e.g., Barr et al. 2012 in Uganda, Pandey et al. 2008 in
        India, Andrabi et al. 2017 in Pakistan, and many others).

Co-Impact evidence review series; Review #1                                                            20
Discussion
        In this review, we have presented various initiatives that have been implemented at large
        scale, usually through government channels, resulting in large learning or completion
        gains, especially for girls. However, the reader may ask: which of these initiatives is the
        best bet? Unfortunately for anyone seeking a silver bullet, the answer comes down to
        the economist’s favorite answer: it depends. Let us demonstrate with two—hopefully
        obvious—examples.
        When is school construction an effective intervention? In Benin in the 1990s, the
        government or NGOs constructed more than 1,500 new schools, and enrollment
        surged by almost 200,000, driven mostly by girls. But the increase in enrollment was
        concentrated in rural areas, where there were fewer schools before the construction
        boom (Deschênes and Hotte 2019). The interventions where school construction was
        effective, in Burkina Faso and Niger and Indonesia, were in locations or times when
        schools were scarcer. Is school construction a good bet? Yes, if there are few schools.
        When is school feeding an effective intervention? In contrast to the examples we
        highlighted above, a large-scale school feeding program in Chile had no impact on
        learning outcomes or grade progression (McEwan 2013). A program in Sri Lanka similarly
        showed no impacts. Why not? Chile had already eliminated extreme malnutrition and
        educational outcomes were relatively strong. Likewise, Sri Lanka already had high rates
        of enrollment (Snilstveit et al. 2016). School feeding is a powerful tool, but only in places
        where children face this particular need. Is school feeding a good bet? Yes, if children are
        malnourished. Ultimately, the most effective intervention in a given location will depend
        on the circumstances of that location.
        Thus, how should an education system decide which interventions to invest in to boost
        girls’ education at scale? We propose three aspects for consideration: constraints to girls’
        education, potential solutions, and program costs.

        Constraints to girls’ education
        First, education systems need to identify their key constraints. What are the weaknesses
        in the education system? What is holding girls’ education back? In recent years, a range of
        diagnostic tools have been deployed that have demonstrated challenges in the education
        system overall, and some can be deployed towards understanding girls’ education.
        Several tools measure student learning, including citizen-led assessments like ASER and
        Uwezo, along with school-based assessments included in the Service Delivery Indicators
        and in national and regional exams (ASER 2021; Uwezo 2021; World Bank 2021). These
        can help identify which regions face the biggest gaps in learning. Likewise, high quality
        gender-disaggregated systems data can track student school completion rates. In terms
        of understanding the reasons behind high student dropout or poor student learning,
        some household surveys (e.g., some of the Demographic and Health Surveys or the
        Living Standards Measurement Surveys) ask directly about reasons that girls drop out of
        school. Furthermore, the Service Delivery Indicators measure the health of the teacher
        workforce, with a focus on skills and absenteeism. If a survey demonstrates that teacher

Co-Impact evidence review series; Review #1                                                             21
You can also read